CN-122017723-A - Method, system, equipment and storage medium for estimating DOA of coherent source

CN122017723ACN 122017723 ACN122017723 ACN 122017723ACN-122017723-A

Abstract

The application relates to a method, a system, equipment and a storage medium for estimating DOA of a coherent source, belonging to the technical field of super-surface signal processing. The method comprises the steps of establishing a quadrature modulation super-surface array model with uniformly distributed super-surface units, performing space-time modulation on an incident signal by using the model to form a single-channel receiving signal, sequentially performing despreading, down-conversion, integral zero clearing and filtering and decoherence processing based on a forward-backward space smoothing technology on the receiving signal to obtain an observation vector, inputting the observation vector into a trained depth expansion ADMM (adaptive modulation) network, sequentially performing sparse feature extraction, physical constraint projection and dual residual updating to obtain a sparse space spectrum vector, mapping the sparse space spectrum vector into a space power spectrum vector, and determining a DOA estimated value by performing spectrum peak search on the space power spectrum vector. The DOA estimation method can realize high-precision and high-efficiency DOA estimation under the condition of low signal-to-noise ratio, and has stronger reliability.

Inventors

SU XIAOLONG
YANG JIACHENG
HU PANHE
Li Zhaojuanhao
HE LIDA
NIU XINYI
LIU ZHEN

Assignees

中国人民解放军国防科技大学

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. A method for estimating the DOA of a coherent source, the method comprising: Establishing a quadrature modulation super-surface array model with uniformly distributed super-surface units, performing space-time modulation on incident signals by using a quadrature modulation coding matrix of each super-surface unit in the model, and performing equal-weight linear superposition on reflection signals of all the super-surface units in space to converge to form single-channel received signals; Sequentially performing despreading, down-conversion, integral zero clearing and filtering and decoherence processing based on a forward-backward space smoothing technology on the received signal to obtain an observation vector of the received signal; Constructing and training a depth expansion ADMM network comprising a sparse feature extraction sublayer, a physical constraint projection sublayer and a dual residual error updating sublayer, wherein network input data is spliced by a complex form observation vector and an array manifold matrix in the training process, and a mixed loss function comprising sparsity auxiliary loss, support set reconstruction loss and physical consistency loss is solved by utilizing a sparse space spectrum vector output by the network so as to perform end-to-end training on the network until the optimal depth expansion ADMM network is converged to obtain, wherein the array manifold matrix is a learnable parameter of the network; And inputting an observation vector into the optimal depth expansion ADMM network, sequentially performing sparse feature extraction, physical constraint projection and dual residual error updating to obtain a sparse spatial spectrum vector, mapping the sparse spatial spectrum vector into a spatial power spectrum vector, searching spectral peaks of the spatial power spectrum vector, and determining DOA estimated values according to positions corresponding to the spectral peaks.
2. The method for estimating DOA of a coherent source according to claim 1, wherein establishing a quadrature modulated super-surface array model with uniformly arranged super-surface elements comprises: Quadrature modulated super surface array model The units of the super surface are uniformly distributed and formed, and the spacing between the adjacent units of the super surface is as follows First, the Received signals of individual super surface units Expressed as: ; In the formula, Is the number of the super-surface unit, In the number of units of the super-surface, In order to take the number of shots in a short time, For the number of incoming signals, For the number of incoming signals, Indicating the wavelength of the incident signal, Represent the first The pitch angle of the individual incident signals, Represent the first White gaussian noise when the individual super surface units are spatially modulated, Represent the first The number of incoming coherent signals is chosen to be, For the center carrier frequency of the system, Is the first Baseband frequency offset of the incident signal, particularly when In the time-course of which the first and second contact surfaces, Is an ideal strong coherent source.
3. The method for estimating DOA as claimed in claim 2, wherein after the incident signals are spatially modulated by the orthogonal modulation and coding matrix of each super-surface unit in the model, the reflected signals of all super-surface units are subjected to equal-weight linear superposition in space, and are converged to form single-channel received signals Expressed as: ; In the formula, Gaussian white noise which is the received signal; is the first Continuous space-time modulated signal of each super-surface unit in whole observation time, which is single period modulated signal Periodic prolongation of (2); a sequence number representing the modulation period, Single period modulated signal ; Is the first The first super-surface unit in one modulation period Modulation code corresponding to the orthogonal modulation code matrix Line 1 The elements of the columns, the quadrature modulation coding matrix of each super-surface unit is Hadamard matrix of order of matrix size ; A rectangular pulse function within a single symbol duration, the specific expression of which is: , For a single complete modulation period, For the duration of a single symbol, For the modulation of the code sequence number, , To be in one modulation period The modulation code length contained therein.
4. A method for estimating a source of coherent information DOA according to claim 3, wherein sequentially despreading, down-converting, integrating clear filtering, and decoherence processing based on forward-backward spatial smoothing technique are performed on the received signal to obtain an observation vector of the received signal, comprising: Reception of signals by means of the mutually orthogonal properties of Hadamard codes Despreading the received signal Multiplying the two orthogonal code sequences of the corresponding channels respectively, multiplying the two orthogonal code sequences by a complex exponential function to perform down-conversion processing, and shifting the down-conversion processing to baseband Baseband signal after demodulation of individual channels Expressed as: ; Wherein, the In order to take the number of shots in a short time, For the center carrier frequency of the system, Is the first Orthogonal modulation coding sequences corresponding to the channels; Introducing an integral clear filter, strictly following a complete modulation period Block reconstruction and mean calculation are carried out on the baseband signal, the first The first channel is at Recovered signal in each modulation period Expressed as: ; Wherein, the As an integral variable, representing the time history within the modulation period; recovering signals using all channels Constructing an array-received raw covariance matrix Expressed as: ; Wherein, the For the number of valid shots after integration, Represents a conjugate transpose; The orthogonal modulation super-surface array model is realized by adopting a forward and backward space smoothing technology The units of the super surface are divided into A plurality of mutually overlapped subarrays, each subarray comprising Each super-surface unit calculates covariance matrix of each subarray and averages to obtain forward smooth matrix The method is specifically expressed as follows: ; Wherein, the Representation from the original covariance matrix The first extracted from Diagonal block matrixes corresponding to the subarrays; Representing a complex field; Using transposition matrices For a pair of Performing conjugate overturning treatment to obtain a backward smooth matrix The method is specifically expressed as follows: ; In the formula, Representing conjugate operation; Synthesizing the forward smoothing matrix and the backward smoothing matrix to obtain a final decorrelated smoothing covariance matrix Expressed as: ; smoothing covariance matrix of decorrelation Vectorization and normalization processing to form observation vector of received signal 。
5. The method of claim 4, wherein the sparse feature extraction sub-layer, the physical constraint projection sub-layer and the dual residual updating sub-layer in the deep expansion ADMM network are respectively expressed as: ; Wherein, the 、 And Respectively represent the network th Layer output sparsified spatial spectral vector, network number Layer auxiliary variable for physical constraint projection, network No The layer is used for updating Lagrangian multiplier directions of dual residual errors; Is a soft threshold operator defined as The method is used for realizing the sparsification and denoising of the features; A feature vector representing an input soft threshold operator, A threshold parameter representing a soft threshold operator; in the form of an array manifold matrix, Is that Is to be used in the present invention, Is an observation vector in the form of a complex number, Is a unitary matrix, network No Punishment parameters of layers Regularization parameters Relaxation factor Globally shared array manifold matrix Are all learnable parameters in the network, and the network initialization state 、 And Zero vector, in the first The layer, the network carries out the mapping of three functional sub-layers in turn, through the maximum layer number of network setting After cascade forward propagation, the network outputs the sparse spatial spectrum vector of the last layer And outputting the optimized array manifold matrix.
6. The method of claim 5, wherein the network input data is obtained by splicing a complex form of observation vector and an array manifold matrix in the training process, and the method comprises the steps of: First define Dictionary matrix composed of virtual guide vectors corresponding to each grid angle under preset grid, wherein the grid angle range is from To the point of The grid interval is Expressed as: ; Wherein, the first At various angles Corresponding virtual steering vectors Composed of Kronecker products of smoothed subarray steering vectors, i.e. In the formula (I), in the formula (II), Represents the Kronecker product of the equation, Is comprised of A smooth subarray steering vector for each of the subsurface units, where The individual elements are defined as: , indicating the wavelength of the incident signal, For the spacing between adjacent super-surface units, Representing conjugate operation; Will be Reconstruction into an array manifold matrix consisting of real and imaginary parts At the same time observe vector Reconstruction into a complex form of observation vectors consisting of a real part and an imaginary part Expressed as: ; ; Wherein, superscript Representing a transpose; Will eventually And The common input depth spreads the ADMM network for end-to-end training.
7. The method of claim 6, wherein solving a hybrid loss function comprising sparsity assistance loss, support set reconstruction loss, and physical consistency loss using sparse spatial spectral vectors of network outputs, comprises: the sparse spatial spectral vector of the network output is expressed as ; Sparsity-assisted loss for constrained output sparsity, expressed as L1 norm loss ; Support set reconstruction loss for measuring sparse spatial spectrum vector and real label of prediction output Is expressed as the mean square error Wherein, the method comprises the steps of, Is the square of the L2 norm; the physical consistency loss is used for measuring the coincidence degree of the reconstruction signal and the actual observation signal, and the sparse space spectrum vector is used for measuring the coincidence degree of the reconstruction signal and the actual observation signal And array manifold matrix Multiplying the signals to reconstruct signals, and then carrying out complex form observation vector input by the network Calculating a mean square error, expressed as ; The final mixing loss function is expressed as: Wherein, the method comprises the steps of, As a result of the sparse weight parameters, Is a physical consistency weight parameter.
8. A coherent source DOA estimation system, the system comprising: The first module is used for establishing a quadrature modulation super-surface array model with uniformly arranged super-surface units, carrying out space-time modulation on incident signals by a quadrature modulation coding matrix of each super-surface unit in the model, and carrying out equal-weight linear superposition on reflection signals of all the super-surface units in space to form a single-channel receiving signal by aggregation; the second module is used for sequentially performing despreading, down-conversion, integral zero clearing and filtering and decoherence processing based on a forward-backward space smoothing technology on the received signal to obtain an observation vector of the received signal; The third module is used for constructing and training a depth expansion ADMM network comprising a sparse feature extraction sublayer, a physical constraint projection sublayer and a dual residual error updating sublayer, wherein network input data are spliced by a complex form observation vector and an array manifold matrix in the training process, and a mixed loss function comprising sparsity auxiliary loss, support set reconstruction loss and physical consistency loss is solved by utilizing a sparse space spectrum vector output by the network so as to perform end-to-end training on the network until the network converges to obtain an optimal depth expansion ADMM network; And a fourth module, configured to input an observation vector into the optimal depth expansion ADMM network, sequentially perform sparse feature extraction, physical constraint projection and dual residual error update, obtain a sparse spatial spectrum vector, map the sparse spatial spectrum vector into a spatial power spectrum vector, and determine a DOA estimation value from a position corresponding to a spectral peak by performing spectral peak search on the spatial power spectrum vector.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Description

Method, system, equipment and storage medium for estimating DOA of coherent source Technical Field The application relates to the technical field of super-surface signal processing, in particular to a method, a system, equipment and a storage medium for estimating DOA of a coherent information source. Background DOA estimation is one of main research directions of array signal processing, and has wide application in the fields of wireless sensing, communication, radar and the like. In recent years, reconfigurable intelligent super-surface (RIS) has received extensive attention due to low cost, flexible deployment and simple design, can accurately regulate and control the amplitude and phase of electromagnetic waves, and provides a new path for DOA estimation. Researchers combine the method with array signal processing, and realize direction finding through a single receiving channel, so that hardware cost and system complexity are greatly reduced. However, the conventional DOA estimation method based on RIS has the defects that the estimation error is large in a low signal-to-noise ratio scene, and a coherent signal source generated by multipath effect can cause rank deficiency of a covariance matrix of received data, so that the traditional algorithm is invalid. In order to solve the problem of estimating rank deficiency of a coherent information source, a compressed sensing reconstruction DOA estimation method based on an alternating direction multiplier method (ALTERNATING DIRECTION METHOD OF MULTIPLIERS, ADMM) is proposed, but the method still has the defects that the number of iteration layers of the ADMM algorithm is large, so that the calculation complexity is high and the real-time requirement is difficult to meet, in addition, a plurality of super parameters are required to be set manually in the algorithm iteration process, the selection of the super parameters lacks a unified optimal criterion, if the parameters are set improperly, the convergence speed of the algorithm is directly influenced, and even the DOA estimation accuracy is reduced. Disclosure of Invention Based on this, it is necessary to provide a method, a system, a device and a storage medium for estimating the DOA of a coherent source, which can realize high-precision and high-efficiency DOA estimation under the condition of low signal-to-noise ratio, and has stronger reliability. A method of coherent source DOA estimation, the method comprising: Establishing a quadrature modulation super-surface array model with uniformly distributed super-surface units, performing space-time modulation on incident signals by using a quadrature modulation coding matrix of each super-surface unit in the model, and performing equal-weight linear superposition on reflection signals of all the super-surface units in space to converge to form single-channel received signals; Sequentially performing despreading, down-conversion, integral zero clearing and filtering and decoherence processing based on a forward-backward space smoothing technology on the received signal to obtain an observation vector of the received signal; Constructing and training a depth expansion ADMM network comprising a sparse feature extraction sublayer, a physical constraint projection sublayer and a dual residual error updating sublayer, wherein network input data is spliced by a complex form observation vector and an array manifold matrix in the training process, and a mixed loss function comprising sparsity auxiliary loss, support set reconstruction loss and physical consistency loss is solved by utilizing a sparse space spectrum vector output by the network so as to perform end-to-end training on the network until the optimal depth expansion ADMM network is converged to obtain, wherein the array manifold matrix is a learnable parameter of the network; the observation vector is input into an optimal depth expansion ADMM network, sparse feature extraction, physical constraint projection and dual residual error updating are sequentially carried out, a sparse space spectrum vector is obtained and mapped into a space power spectrum vector, spectrum peak searching is carried out on the space power spectrum vector, and DOA estimated values are determined according to positions corresponding to spectrum peaks. In one embodiment, establishing a quadrature modulated super surface array model with uniformly arranged super surface units includes: Quadrature modulated super surface array model The units of the super surface are uniformly distributed and formed, and the spacing between the adjacent units of the super surface is as followsFirst, theReceived signals of individual super surface unitsExpressed as: ; In the formula, Is the number of the super-surface unit,In the number of units of the super-surface,In order to take the number of shots in a short time,For the number of incoming signals,For the number of incoming signals,Indicating the wavelength of the incident signal,Represent t